1

Estimation of VOC Emission Factors from Flux Measurements using 1

a Receptor Model and Footprint Analysis 2

3

Sri Harsha Kota1, Changhyoun Park

2, Martin C. Hale

2, Nicholas D. Werner

2, Gunnar W. Schade

2, 4

and Qi Ying1,*

5

1Zachry Department of Civil Engineering, Texas A&M University, 6

College Station, Texas, USA 7 2Department of Atmospheric Sciences, Texas A&M University, 8

College Station, Texas, USA 9

10

Abstract 11

Fluxes of 18 volatile organic compounds (VOCs) collected during May to July 2008 from a tow-12

er platform 60 m above the surface in an urban Houston residential area were analyzed using re-13

ceptor-oriented statistical models and an analytical flux-footprint model to resolve daytime 14

source specific emissions rates. The Multilinear Engine version 2 (ME-2) was used to determine 15

that five sources were responsible for the measured flux at the tower: (i) vehicle exhaust, (ii) a 16

foam plastics industrial source with significant pentane emissions, (iii) consumer and commer-17

cial solvent use emissions, (iv) a biogenic emissions source dominated by isoprene, and, (v) 18

evaporative fuel emissions. The estimated median daytime (0700-1900 CST) hourly emission 19

rate from the foam plastics industry was 15.7±3.1 kg h-1

, somewhat higher than its permitted 20

hourly emission rates. The median daytime vehicle exhaust VOC emission rate of 14.5±2 g h-1

21

vehicle-1

, was slightly higher than our estimation using the Motor Vehicle Emission Simulator 22

(MOVES) with a county-representative vehicle fleet of year 2008 (11.6±0.2 g h-1

vehicle-1

). The 23

median daytime evaporative fuel VOCs emission rate from parked vehicles was 2.3±1.0 g h-1

24

vehicle-1

, which is higher than MOVES estimations and could not be explained by the age of the 25

vehicle fleet, indicating either locally higher evaporative emission sources in the footprint or an 26

underestimation of evaporative emissions by MOVES, or both. 27

28

Keywords: Source Apportionment; Positive Matrix Factorization (PMF); Multilinear Engine 29

(ME2); Motor Vehicle Emission Simulator (MOVES); Vehicle Exhaust; Evaporative Emissions; 30

Foam Plastics Industry. 31

*Corresponding Author. Email: qying@civil.tamu.edu. Phone: (979) 845-9709. Fax: (979) 862-1542.

2

1. Introduction 32

Volatile organic compounds (VOCs) play a prominent role in photochemical reactions that lead 33

to the formation of ozone and secondary particulate matter, thus directly affect regional air quali-34

ty and global climate (Atkinson, 2000; Kroll and Seinfeld, 2008). A number of VOCs are also 35

classified as hazardous air pollutants by the US EPA due to their adverse health effects. Alt-36

hough significant efforts have been devoted in the past to develop and improve VOC emission 37

inventories, large uncertainties and biases remain (Brown et al., 2004; Buzcu and Fraser, 2006; 38

Reid et al., 2000). Under-reported or unreported anthropogenic emissions in VOC emission in-39

ventories are one of the major factors that affect air quality models and forecasts, particularly for 40

ozone (McKeen et al., 2009; Nam et al., 2006) in metropolitan areas. In urban areas, vehicle 41

emissions account for a large fraction (e.g. approximately 30% in the Houston metropolitan area) 42

of the anthropogenic VOC emissions (Ying and Krishnan, 2010), but the accuracy of the emis-43

sions depends largely on the vehicle emission factor models used in the estimations. The uncer-44

tainty in the VOC emission inventory can thus significantly affect the evaluations of VOC emis-45

sions on air quality, human health, climate and the design of effective control strategies to miti-46

gate adverse effects. 47

Ozone and particulate air quality in Houston, the 4th

largest metropolitan area in the United 48

States (US) with a population of over 2.2 million, is significantly influenced by the VOC emis-49

sions from petrochemical, industrial and motor vehicle sources (Gilman et al., 2009; Kim et al., 50

2011; Vizuete et al., 2008; Ying and Krishnan, 2010). This complex mixture of VOCs has led to 51

a number of studies to quantify the contributions of different sources to the observed VOC con-52

centrations in the area. Fujita et al. (1995) used Chemical Mass Balance (CMB) modeling to 53

study the VOC data from Photochemical Assessment Monitoring Stations (PAMS) and conclud-54

ed that refineries are the dominant VOC source in the Houston Ship Channel (HSC). Henry et al. 55

(1997) used a multivariate receptor model to study data collected during the Coastal Oxidant As-56

sessment for Southeast Texas (COAST) study, and showed that self-reported emissions by vari-57

ous industries in the HSC area were unreliable. Kim et al. (2005), Buzcu and Fraser (2006), and 58

Luchner and Rappenglück (2010) applied the Positive Matrix Factorization (PMF) technique to 59

study VOC sources in Houston, and concluded that refineries, petrochemical industries, vehicle 60

emissions and biogenic sources are all important contributors to the ambient VOC abundance. 61

While these receptor-oriented source apportionment studies are useful in understanding sources 62

3

of VOCs in the atmosphere, the calculations are based on measured VOC concentrations and 63

thus are not directly related to the actual emission rates of VOCs from various sources. 64

Instead of using ambient concentrations, VOC fluxes calculated from micrometeorological 65

and VOC gradient measurements have been used in the past to estimate biogenic emission fluxes 66

from forest areas (Karl et al., 2001; Langford et al., 2010a; Lee et al., 2005; Spirig et al., 2005). 67

Recently, micrometeorological flux measurements have also been applied to measure emission 68

rates of anthropogenic and biogenic VOCs in urban environments (Karl et al., 2009; Langford et 69

al., 2009; Park et al., 2010; Velasco et al., 2009). The urban fluxes, usually measured from tall 70

towers, are used to directly infer the emission rates of pollutants from upwind areas using foot-71

print models (Langford et al., 2010b). The results of footprint modeling, resulting in apparent 72

surface fluxes, can be used in conjunction with an analysis of land use/land cover and/or traffic 73

count data to infer specific emission rates for different sources included in the footprint areas 74

(Park et al., 2011). Although this technique is useful, it is not straightforward to identify respon-75

sible sources within the footprint area due to the high spatial heterogeneity of emission sources 76

in typical urban environments. In addition, different sources are likely responsible for different 77

groups of VOCs, while a typical footprint analysis applies to a homogeneous source distribution. 78

In this study, simultaneous fluxes of 18 VOCs are used in receptor-oriented statistical anal-79

yses to resolve sources of measured VOC fluxes in an urban environment. Results of the source 80

attribution analysis are used alongside flux-footprint modeling to determine the emission rates of 81

VOCs from the different sources. To the knowledge of the authors, this is the first time such a 82

combination of VOC flux measurements and receptor-oriented source apportionment analyses is 83

applied to resolve source specific emission rates of VOCs in urban locations. 84

85

2. Methodology 86

2.1 Data and data uncertainties 87

The experimental setup and data collection have been described in detail in Park et al. (2010) and 88

are only briefly summarized here: Meteorology data and concentration and flux of 18 VOCs (see 89

Table 1 for the list of the VOCs) were measured at 60 m above ground level (agl) from a tall 90

communication tower owned by the Greater Houston Transportation Company (hereinafter re-91

ferred as the Yellow Cab Tower, or YCT) in an older neighborhood 3-4 km north of downtown 92

Houston (Northside Village area; Figure S3) from May 23 to July 27, 2008. The area surround-93

4

ing the tower is mostly residential, with several multi-lane roadways and a light industrial area 94

surrounding YCT. It was estimated that 29% of the surrounding areas are covered by trees and 95

shrubs, among which a mature oak tree population results in significant emissions of isoprene as 96

reported in Park et al. (2011). Figure 1 shows a schematic view of the area surrounding the flux 97

measurement site, including the locations of YCT, major through-traffic roadways, two YC park-98

ing lots, and some other potentially contributing sources that are mentioned in this manuscript. 99

The VOC concentrations were measured using a dual channel gas chromatograph with flame 100

ionization detectors (GC-FID) and the fluxes were determined using a relaxed eddy accumula-101

tion (REA) setup. In summary, the REA setup measures the concentration of a VOC species in 102

atmospheric updrafts (Cup) and downdrafts (Cdown) over an averaging time period of 30 minutes. 103

The resultant flux (F) is calculated using Equation (1): 104

(1)

where β is a flux correction factor (in this study β=0.335, see Park et al. (2011) for details) and 105

σw is the standard deviation of vertical wind speed of each 30 min sampling period. At the top of 106

each hour a 30-minute sample was taken and it was assumed to represent the average flux of that 107

hour. Species specific method detection limit (MDL) of the concentration measurements is in-108

cluded in Table 1. Flux MDL (species and sample specific) was based on regular (every 30th

run) 109

GC-FID channel intercomparisons by obtaining identical air samples into the Teflon bags. The 110

95% confidence limit (95%CI) of the difference in concentration between these samples (exclud-111

ing outliers) was used to calculate sample specific flux MDLi,j (MDL for the ith

sample and jth

112

species), as shown in Equation (2): 113

MDLi,j= β σw,i ×(95%CI)j (2)

Prescreening was performed to eliminate data obtained under low turbulence and non-114

stationary flux conditions: Flux and concentration data obtained under friction velocities, u*<0.2 115

m s-1

or 60-m agl wind speed less than 2 m s-1

were not retained (Park et al., 2011). This particu-116

larly reduced the amount of nighttime data, when turbulence was weaker. Furthermore, periods 117

with questionable stationarity of high frequency CO2 and low frequency CO data, following the 118

standard deviation technique used by Foken and Wichura (1996), were removed. For the remain-119

ing samples (760 30-minute samples), data below MDLi,j were replaced with 0.5MDLi,j, and 120

missing data were set to the median concentration of the species (Polissar et al., 1998). A species 121

was marked as missing in the flux data set if both of the two GC-FID channels was missing or 122

F

wC

upC

down

5

had concentrations below MDL. If only one of the channel was missing or below MDL, the flux 123

was taken as 0.5MDLi,j. 124

The uncertainty (σ) associated with each sample for receptor-oriented source apportionment 125

analyses was estimated using Equation (3a) when the concentration or absolute value of flux was 126

less than or equal to MDLi,j or otherwise using Equation (3b) (Polissar et al., 2001). 127

, ,

1 1

2 3ji j i jMDL MDL

(3a)

i, j

i, j

2 MDLi, j

2

(3b)

where μi,j is the absolute analytical uncertainty of the measured concentration or flux. A relative 128

precision of 10% for concentrations based on internal standard variability was used to calculate 129

the absolute uncertainty for the concentration data (see Park et al., 2010 for further details). Ab-130

solute analytical uncertainty of fluxes was calculated through propagation of uncertainties in the 131

up and down draft concentrations, as shown in Equation (4): 132

22 22, , , ,

, 2

, , , ,

up i j down i j wi j

wup i j down i j

C CC

C C C

(4)

where the relative uncertainty of concentration (ΔC/C) in up and down drafts was taken as 10% 133

and relative uncertainty of σw (∆σw/σw) was taken as 5%. σi,j for missing data was set to five times 134

the median concentration. 135

136

2.2 Source apportionment of fluxes 137

The Multilinear Engine version 2 (ME-2) (Paatero, 1999, 2004), the underlying solver for the 138

United States Environmental Protection Agency Positive Matrix Factorization (PMF) 3.0 model 139

(USEPA, 2008; downloaded from http://www.epa.gov/heasd/products/pmf/pmf.html), was used 140

to solve the source apportionment problem for both concentration and flux data. ME-2 can be 141

used to solve the least-square problem from many types of factor analysis (including PMF) and 142

has been applied before in a number of air pollution source apportionment studies (Amato et al., 143

2009; Kim et al., 2003; Ramadan et al., 2003; Wu et al., 2007). One of the features of ME-2 that 144

is useful for this study is that it can be configured to allow negative source contributions (Norris, 145

2009), which is natural for the flux data as the measured net flux can be a superposition of gross 146

positive and negative fluxes from different sources and sinks. In this study, for the concentration 147

6

data source apportionment, ME-2 was configured to use non-negativity constrains for source 148

profiles and contribution matrices, and to allow an unconstrained solution for the contribution 149

matrices when it was applied to determine source contributions for the flux data. For both con-150

centration and flux source apportionment, 100 bootstrap runs with a block size of 16 and a min-151

imum correlation R-value of 0.6 were conducted to ensure proper solutions and to estimate the 152

uncertainties to the estimated profiles. 153

154

2.3 Automatic identification of source profiles 155

To attribute the ME-2 resolved source profiles (mg mg-1

) to a specific source, the profiles were 156

compared with renormalized VOC profiles (including only the 18 species analyzed in this study) 157

from the SPECIATE 4.2 database (Hsu and Divita, 2008), a VOC and PM speciation profile data 158

base maintained by the US EPA, using Equation (5): 159

(5)

where fi and si are the ith

matching component in the ME-2 resolved source profile and the SPE-160

CIATE 4.2 profile, respectively. θ is bounded between 0 and 1, where 1 indicates perfect agree-161

ment. Top 20 matching SPECIATE profiles were then manually checked to determine the source 162

type for the ME-2 source profile. 163

164

2.4 Emission rate estimation 165

The source-apportioned VOC fluxes at YCT were used to estimate the VOC emission rates of 166

the identified sources using the analytical footprint model described by Kormann and Meixner 167

(2001). Generally, the flux measured at a certain height, F(0,0,zm), can be related to upwind sur-168

face fluxes, as described by equation (6), 169

0(0,0, ) ( , ,0) ( , , )m mF z F x y x y z dxdy

(6)

where the two-dimensional flux-footprint probability density function represents the probabil-170

ity of a unit flux at (x,y,0) that reaches the flux measuring location (0,0,zm). The Kormann and 171

Meixner model assumes a homogeneous underlying surface and well-defined atmospheric turbu-172

lence regimes. It is attractive due to the limited amount of input parameters required, providing a 173

symmetric flux footprint function with results similar to a more sophisticated model (Kljun et al., 174

1

2 2

1 1

N

i ii

N N

i ii i

f s

f s

7

2004; Kljun et al., 2002). However, it has not yet been rigorously tested in a turbulently more 175

complex urban environment, which is heterogeneous both in terms of roughness length (due to 176

different building and vegetation heights) and heat flux/stability. In addition, the analytical foot-177

print model strictly only provides the flux footprint function at the displacement height, which 178

may vary between 5 and 13 m at this site. 179

The flux footprint model output option of EdiRe flux processing software 180

(http://www.geos.ed.ac.uk/abs/research/micromet/EdiRe/) was used to calculate the 2D gridded 181

flux-footprint probability function φ (i.e. integrated within each grid cell) in a domain of 6×6 182

km2, i.e. using a square grid of 30 m, with the YCT at the center of the domain. Hourly flux 183

footprint probability values used in the following analyses were calculated by averaging the 30-184

min footprint values within a given hour. If flux F(0,0,zm) is known, the calculated footprint 185

probabilities can be used to estimate the surface emission fluxes by inverting a discrete form of 186

Equation (5). For the emission rate analysis in Section 3.2, individual periods were removed 187

from the analysis if the domain sum of the flux probability φ was less than 0.7 to ensure that a 188

sufficient amount of the flux footprint lies within the computation domain. Nighttime data (2000 189

– 0600 CST) were completely excluded to further reduce uncertainty. This resulted in a total re-190

moval of 363 samples before source specific criteria were specified for the emission rate analysis. 191

192

3. Results and Discussion 193

Analysis of the concentration data are described in greater detail in the Supplementary Materials 194

(Figure S1-S2). In summary, measured concentrations were generally well-reproduced by ME-2 195

with five factors, representing consumer and commercial solvent use emissions, an industrial 196

source dominated by pentane emissions (referred to as the “foam plastics industry” emissions 197

hereafter), vehicle exhaust, evaporative emissions and a biogenic emissions source. While this is 198

an expected result in line with previous work, the additional flux dimension can provide further 199

insight, wherefore in the following analyses we focus on the flux data. A comparison of the 200

source profiles and relative source contributions derived from concentration and flux data can be 201

found in Figure S3 and Table S6. 202

203

3.1 Source apportionment of flux data 204

8

Results generated by ME-2 assuming 4, 5 and 6 factors (sources) were explored. The value of the 205

sum-of-squares objective function Q to its expected (or theoretical) value Qexpected, Q/Qexpected, for 206

4, 5 and 6 factors were 1.4, 1.15 and 1.1, respectively. The 5 and 6 factor solutions had similar 207

correlation coefficients (r2) for the total VOC mass (0.935 vs. 0.941), and the amount of total 208

VOC represented (84% vs. 85%). However, the 6-factor solution resulted in two very similar 209

consumer and commercial solvent use emission factors. Similar results were also achieved for 210

ME-2 analysis of the concentration data (see Supplementary Materials). Thus, the 5-factor solu-211

tion, with rotational parameter FPEAK of 4 (Norris, 2009) (see Tables S2 and S3 for more in-212

formation), was used in the following analyses. 213

Figure 2 shows the predicted source profiles, which are determined to represent (1) consumer 214

and commercial solvent use emissions, with C4 as the dominant species (43% of the VOCs in the 215

profile, θ=97%), (2) a foam plastics industry emissions source, with pentane as the dominant 216

species (60% of the VOCs in the profile), (3) vehicle tailpipe exhaust emissions, dominated by 217

TOLU and MPXYL (19% and 24% of the VOCs in the profile, θ=92%), (4) evaporative emis-218

sions, dominated by IC5H12 (30% of the VOCs in the profile, θ=94%), and (5) biogenic emis-219

sions, dominated by isoprene (44% of the VOCs in the profile, θ=91%). Table S4 in the supple-220

mentary materials lists the top matching profiles in the SPECIATE 4.2 database. Figure 2(a) 221

shows the consumer and commercial solvent use emissions factor, which apparently does not 222

represent fugitive evaporative fuel emissions from vehicles because butane, primarily used as 223

aerosol propellant, is the dominate species in that profile, and it does not have a significant con-224

tribution from IC5H12 (less than 5%). Rubin et al. (2006) reported that IC5H12 (26.6%) is much 225

more important than n-butane (8.0%) among the most abundant components in evaporative fuel 226

emissions, which agrees much better with the profile shown in Figure 2(d). The profile shown in 227

Figure 2(a) more closely resembles several consumer and commercial profiles in the SPECIATE 228

4.2 data base than the closest vehicle fuel evaporation profiles (θ=93%). Figure 2(b) was deter-229

mined to be a foam plastics industry source based on the directional dependence of the factor 230

(Figure 5) and a survey of the surrounding area as described later in this section. Figure S3 231

shows a comparison of the profiles based on concentration and flux data. The two profiles are 232

very similar and no consistent trend of the major species could be found. 233

Figure 3 shows that ME-2 predicted VOC fluxes were lower than observations and for a few 234

species such as MACR, MVK and MEK, the underestimations were more significant. Normal-235

9

ized mean bias factor (NMBF) and normalized mean absolute error factor (NMAEF) for the spe-236

cies were in the range of -0.37 to -0.06 and 0.43 to 0.75 for most species, as shown in Table S5. 237

These two statistical measures were recommended for datasets with both positive and negative 238

values (Gustafson and Yu, 2012). Definitions of NMBF and NMAEF can be found in the Sup-239

plementary Material. MACR and MVK are oxidation products of isoprene, and MEK is an oxi-240

dation product of n-butane and isopentane. The receptor-oriented statistical methods typically do 241

not work as well for these compounds because the ratio of these products to their precursor 242

changes as they are transported towards the receptor. 243

The diurnal variation of the contribution of each source to the measured total VOC flux is 244

presented in Figure 4. Figure 4(a) shows that VOC flux due to consumer and commercial solvent 245

use emissions is higher during the day. Contributions at late night until early morning were com-246

paratively smaller, possibly due to lower temperature as well as decreased activities associated 247

with the emissions of these VOCs. Figure 4(b) indicates a clear diurnal pattern of contributions 248

from the foam plastics industry source, with an interquartile range (IQR) of approximately; 0-2 249

mg m-2

h-1

during the late afternoon, compared to the IQR of approximately 0-1 mg m-2

h-1

in 250

morning hours. Contributions were lower during the night as emissions of this source are also 251

largely driven by ambient temperatures and work activity. Figure 4(c) shows a clearly higher 252

daytime than nighttime contribution from vehicle exhaust with a clear morning rush hours peak 253

around 0700-0900 CST (IQR 0-2 mg m-2

h-1

), while the IQRs of surrounding hours’ (0600-0700 254

and 0900-1000 CST) fluxes are approximately 0-1 mg m-2

h-1

. The morning peak coincided with 255

the weekday rush hours observed on Hardy St., the nearest major thoroughfare near the sampling 256

site (See Table S7). More discussions of the nearby roadways can be found in Section 3.2.2. Fig-257

ure 4(d) indicates that contributions from evaporative emissions were similar to vehicle exhaust 258

at the YCT with narrower IQR and the factor did not display an as significant rush hour peak 259

signature as observed in Figure 4(c). This suggests that the evaporative VOC emissions source 260

was likely not dominated by running losses from vehicles on the nearby roadways. Lastly, Figure 261

4(e) indicates that the highest biogenic isoprene contributions from the surrounding oak tree 262

population occurred in the early afternoon due to an optimum radiation and temperature envi-263

ronment at that time of day (Park et al., 2011). Biogenic isoprene contributions peaked at 1200 to 264

1400 CST, with a maximum median flux of 2.1 mg m-2

h-1

. 265

10

Figure 5 shows the wind directional dependence of the fluxes from the different resolved 266

sources. Approximately 75% of the time during the analysis, the receptor was under the influ-267

ence of southerly winds. Figure 5(a) indicates that most of the contributions from consumer and 268

commercial solvent use source were from south to southeast directions of the tower without a 269

strong directional dependence. This suggests that this is a regional source rather than a collection 270

of a few point sources. In contrast, the contribution of the foam plastics industry emissions 271

source, Figure 5(b), is almost exclusively from the south-southeast direction throughout the day. 272

This strong wind direction dependence suggests contributions from a well-defined source. A sur-273

vey of the area southeast of YCT revealed this potential source to be an industry specializing in 274

foam plastics approximately 1.6 km southeast of the receptor location (Figure 1), noting that pen-275

tane is used as an expansion agent in many foam plastics industries (Mills, 2007). As the source 276

is relatively far away from YCT, unidentified non-stationary flux conditions could explain the 277

occasional negative fluxes shown in the figure. 278

The vehicle exhaust contributions, depicted in Figure 5(c), are from all directions as the tow-279

er is surrounded by roadways. However, two directions stand out slightly: Southeast, likely due 280

to optimum overlap of the flux footprint with the major thoroughfare Hardy/Elysian roads, and 281

south-southwest due to significant traffic surrounding two major schools in that direction in ap-282

proximately 1 km distance from YCT. As the sampling site is amidst the parking lots of the Yel-283

low Cab Co., which operates around 1400 vehicles (Mike Spears, Houston Yellow Cab Co., per-284

sonal communication, May, 2011), many of which are parked at different directions from the 285

tower, observed contributions of evaporative emissions from all directions, Figure 5(d), can be 286

expected. Comparison of Figures 5(c) and 5(d) indicates that evaporative emissions contributions 287

followed a different wind direction pattern compared to vehicle exhaust, which supports the ear-288

lier discussion that evaporated gasoline from parked vehicles rather than running vehicles are 289

responsible for most of the observed fluxes. This presumption of attributing evaporative emis-290

sions dominantly to parked vehicles will be discussed in detail in later sections. Lastly, the pres-291

ence of oak trees in the surrounding neighborhood resulted in contributions of biogenic emis-292

sions from all directions, shown in Figure 5(e). Details have been published by Park et al. (2011). 293

Negative fluxes due to biogenic emissions occurred mostly during nighttime, with a strong direc-294

tion dependence pointing towards the HSC area, again suggesting a contribution from non-295

stationary conditions due to emissions from sources advected from outside the footprint domain. 296

11

297

3.2 VOC emission rates using footprint analysis 298

Figure 6 shows the gridded flux footprint probability φ between hours 0700-1200 (374 individual 299

½ hours) and 1300-1900 CST (678 individual ½ hours). During both morning and afternoon 300

hours, the dominant wind direction was southerly so that high values of the footprint function 301

occurred in that direction. Thus, sources to the south of the tower contributed most to the fluxes 302

measured at the tower. Maximum footprint values generally occurred close to YCT and de-303

creased rapidly towards the border of the domain. This suggests that, on average, the footprint 304

model area is sufficient to include the influence of major sources. In the following sections, 305

emission rates for the foam plastics industry, vehicle exhaust and evaporative fuel emissions are 306

discussed. The calculation of isoprene emission factors was discussed by Park et al. (2011) and is 307

not repeated here. Emission rates of VOCs from consumer and commercial solvent use cannot be 308

directly estimated as the unit area emission rate from residential and commercial areas are differ-309

ent. An optimization step is needed to estimate the emission rates for those two different sources. 310

Many factors also affect the unit area emission rates such as the type of commercial facilities and 311

products. Due to limitations of manuscript length, an analysis of the VOC emission rates from 312

this source will be discussed elsewhere. 313

314

3.2.1 Foam plastics industry emissions 315

To narrow the VOC emission rates from the foam plastics industry, only the data with wind di-316

rections between 100 and 170 degrees to the receptor location were considered in the analysis. 317

This resulted in 123 samples which had non-negative flux contributions. Equation (7) was used 318

to calculate the VOC emission factor of the foam plastics industry factor: 319

61

10 Ipentane pentane IE S F

(7)

where Epentane is the VOC emission rate from the industrial source (kg h-1

); SI is area (m2) of the 320

industrial source region, from which VOC emissions are released into the atmosphere; Fpentane is 321

the ME-2-resolved VOC flux (mg m-2

h-1

) for the industry; φ is the flux footprint probability at 322

each grid cell; 10-6

converts the units from mg to kg; and the summation means summing the φ 323

values for the grid cells within the industrial source region. 324

12

In order to estimate the uncertainty of the VOC emissions rate due to uncertainty in φ, a 325

Monte Carlo technique was used with the underlying assumption that uncertainties in φ can be 326

represented by the variation of φ in nearby grid cells. For each valid foam plastics industry emis-327

sions’ flux data point, 4000 simulations were carried out. In each simulation, the emission source 328

region was randomly moved around a fixed center location, assuming a normal distribution with 329

a standard deviation of ±3 grid cells (i.e. ±90 m) in both x and y directions. The average VOC 330

emissions rate for the pentane source for a flux data point was then calculated from these 4000 331

simulations. The number of necessary simulations was determined by incrementally increasing 332

the number of simulations until the mean and standard deviation of the emission rate no longer 333

changed. The number of grid cells that cover the emission source region and the shape of the 334

source region remained constant in these simulations. Once mean emission rates of all data 335

points were determined, extremely high and low hourly mean emission rates within the dataset 336

were removed as outliers, which are defined as samples that fall outside 1.5 times the inter-337

quartile range of the data (Moore and McCabe, 1999). This resulted in removal of 19 samples 338

(15%) from overall emission factor analysis. This outlier removal procedure was also carried out 339

for the analysis of other sources described in subsequent sections. 340

Half of the estimated emission factors had a relative standard deviation of less than 9.6% and 341

90% of the data had a relative standard deviation of less than 19.3%. This suggests that for most 342

of the data points, the uncertainty in the individual emission rate due to uncertainty in φ was 343

quite small. Statistical analysis of the data (Table 2) showed a median of 15.7 kg h-1

with 95% 344

confidence intervals [12.6, 18.8] kg h-1

, and a mean emission rate of 18.5 kg h-1

. Although the 345

area of the source region is related with the plume size of emissions at the displacement height, 346

which is unknown, it is not going to greatly affect Epentane because when SI decreases I also 347

decreases. As long as there is a weak gradient of φ near the source region, Epentane will remain 348

relatively constant. To verify this, the area of the source region was varied from 1 to 15 grid cells 349

in a series of calculations similar to the approach described above but without using the Monte 350

Carlo technique that varies the center of the source region. The resulting mean emission rate var-351

ied slightly between 17.1 and 18.7 kg h-1

. The average of the mean emission rates was 17.7 kg h-1

, 352

which is very similar to the mean using 15 grid cells as the source area. 353

The company has a permit to emit 10.5 kg h-1

, with 45% of emissions from storage. Thus, the 354

estimated mean and median emission factors were 40-60% higher than permitted emissions. 355

13

However, summer, particularly June 2008 had above normal temperatures possibly enhancing 356

emissions, and as comparisons of the Kormann and Meixner model to a more sophisticated back-357

trajectory footprint model suggest a systematically longer “tail” of the Kormann and Meixner 358

model footprint function (Kljun et al., 2003), we cannot exclude a slight high bias in our estimate 359

for this distant source. 360

361

3.2.2 Vehicle exhaust 362

VOC emission factors for vehicle exhaust were also estimated using resolved vehicle exhaust 363

flux and the footprint model. Hourly vehicle volume (Tables S7 and S8) and speed (Tables S9 364

and S10) data were collected on four major roadways for through-traffic (Hardy St., Elysian St., 365

Collingsworth St. and Quitman St.; Figure 1 and Figure S3) near YCT during March and No-366

vember, 2011. Hardy and Elysian are north-south oriented multi-lane roadways one and two 367

blocks east of YCT, respectively. Quitman and Collingsworth are normal two-lane east-west ori-368

ented streets, 7 blocks south, and 4 blocks north of YCT, respectively. In the emission factor cal-369

culation, nearby local streets within approximately 250 m of YCT were also included. Traffic 370

data collected on Hays Street, which is the local east-west oriented street approximately 20 m 371

north of the tower, were assumed to represent general traffic conditions in the surrounding local 372

streets. The names and the locations of the eight nearby local streets included in the emission 373

factor calculation are shown in Figure S4. As shown in Figure S5, these eight local streets and 374

four thoroughfares encompass areas with significant footprint probability. Including additional 375

local roadways further away from YCT is not expected to affect the estimated emission factor. 376

Hourly traffic density (number of driving vehicles per grid cell) for a typical weekday and 377

weekend day, which is needed for the emission factor calculation, was calculated using the col-378

lected traffic data. Although traffic count data were not directly available for the current model-379

ing period, it was assumed that traffic density did not change significantly within a few years in 380

this relatively old neighborhood. This assumption is supported by a less than 1% change in annu-381

al average diurnal traffic (AADT) during 2008-2011 on the freeways surrounding the tower 382

(http://www.txdot.gov/apps/statewide_mapping/StatewidePlanningMap.html). It was further as-383

sumed that vehicle density was uniform at the footprint grid cells of the same roadway, and that 384

the vehicle fleet composition was the same everywhere in the domain so that the VOC emission 385

factor is uniform throughout the domain. 386

14

The hourly emission factors were calculated using Equation (8) based on the ME-2-387

apportioned vehicle VOC exhaust fluxes and the corresponding footprint values, 388

1

3

, , ,

1 1

10yx

NNroad

exhaust i j i j iexhaus

i

t j

j

EF F k

(8)

where EFexhaust is the VOC emission factor for an average vehicle in the domain (g h-1

vehicle-1

), 389

Fexhaust is the ME-2 apportioned vehicle VOC flux at YCT (mg m-2

h-1

) for a specific hour; ζroad

is 390

the roadway mask function, which returns unity if a footprint grid cell belongs to one of the 391

roadways otherwise it returns zero; k is the vehicle density of a grid cell (number of vehicles m-2

); 392

φ is the average hourly footprint value at each grid cell; and 10-3

converts units to g h-1

vehicle-1

. 393

Since the roadways and larger vehicle densities are located dominantly in the east and south di-394

rections of the tower, only hourly data with wind direction between 20 and 270 degrees were 395

considered for the analysis. This resulted in 233 samples which had non-negative flux contribu-396

tions from vehicle exhaust. The total emission rates of the 18 measured VOCs were converted to 397

the total VOC emission rates using a weighing factor of 0.41±0.11 (mass of 18 measured 398

VOCs/total of all VOC mass in a VOC speciation profile) based on the vehicle exhaust profiles 399

available in the SPECIATE 4.2 chemical speciation data base (Hsu and Divita, 2008). 400

Similar to the foam plastics industry emission factor calculations, uncertainty in the vehicle 401

exhaust due to k and φ was estimated using the Monte Carlo technique. For each data point, 402

20000 simulations were carried out. In each simulation, the vehicle density was calculated by 403

randomly varying the vehicle speed and traffic volume based on normal distributions with mean 404

and standard deviations shown in Tables S7-S10. Uncertainty in the footprint function was again 405

estimated by randomly selecting φ values from grid cells with a normal distribution centered at 406

the road grid points and a standard deviation of 3 grid cells in both x and y directions. 29 samples 407

(approximately 12%) were removed as outliers from the analysis. 408

Half of the estimated emission factors had a relative standard deviation of less than 30% and 409

90% of the data had a relative standard deviation of less than 200%. Data points with larger un-410

certainties typically occurred when the wind was not from the south. The gradient of φ is signifi-411

cant near the north-south streets of Hardy and Elysian (Figure 6), which explains some of the 412

higher uncertainty in the estimated hourly emission rates. Table 2 shows the statistical analysis 413

of the hourly data. The mean vehicle exhaust emission rate was 17.5 g h-1

vehicle-1

. The median 414

emission rate was 14.5 g h-1

vehicle -1

, with a 95% confidence interval [12.5, 16.5] g h-1

vehicle-1

. 415

15

The calculated VOC emission factor from an average vehicle was next compared with the 416

emission factors estimated by the Motor Vehicle Emission Simulator (MOVES) model. Calculat-417

ing emission factors for an average vehicle using MOVES and traffic monitoring data has been 418

described in detail in a separate manuscript (Kota et al., 2013). Figure 7 shows the MOVES es-419

timated emission factor for fleet years 2000 to 2008 and their comparison with the flux-footprint 420

estimated emission factors. Variations in the predicted MOVES emission factors were smaller 421

because, e.g., no footprint function is involved in the MOVES emission calculations. The 422

MOVES emission factors for fleet year 2005 to 2008 were within the 95% confidence interval of 423

the median emission factors estimated by the flux-footprint analysis, and the 2005 MOVES 424

emission factor (14.1±0.2 g h-1

vehicle-1

) was closest to the median value (14.5 g h-1

vehicle-1

). 425

However, the mean emission rate from the flux-footprint analysis (17.5 g h-1

vehicle-1

) was clos-426

est to the MOVES estimated emissions for fleet year 2004 (16.5 g h-1

vehicle-1

). 427

In addition to the major roads discussed in this study, two major freeways, US-59 and I-45 428

(both oriented roughly in north-south direction) are located at the edge of the footprint region 429

1270 m toward the east and 1740 m toward the west from the tower, respectively. I-45 and US-430

59 have 35.4 and 22.2 times higher vehicle traffic than Quitman 431

(http://ttihouston.tamu.edu/hgac/trafficcountmap/). Despite these substantially higher traffic vol-432

umes, including the two freeways in the emission factor calculation resulted in only a 2.7% de-433

crease in the average emission factor. 434

435

3.2.3 Evaporative fuel emissions 436

Although there are many potential sources that can contribute to evaporative fuel emissions, we 437

are interested in limiting our analysis to the VOC evaporative flux from stopped and parked ve-438

hicles, mostly at or near YCT and its two major parking lots, located at 90-180 m southeast and 439

60-120 m northeast of YCT (Figure 1). The number of parked taxi cars near the tower was esti-440

mated by counting the number of designated parking spots near the facility (240 and 120 vehi-441

cles in southeast and northeast parking lots, respectively). Since not all the parking spaces were 442

occupied by vehicles, this may give a lower bound estimation of the actual evaporative emission 443

rate. Parked vehicles on the streets in other grid cells can also contribute to the measured flux at 444

the tower. The parked vehicle density in other grid cells was estimated to be 2 vehicles per grid 445

cell based on Google Earth images for the year 2008. To reduce the uncertainty in the emission 446

16

rates estimation, only samples with wind directions between 100-190 degrees (the southeast 447

parking lot) and 20-80 degrees (the northeast parking lot) were included in the calculation. This 448

resulted in 70 samples which had non-negative flux contributions from evaporative emissions. 449

Contributions to evaporative fuel emissions from a gasoline transport company approximately 450

1.2 km to the east of the sampling site (Figure 1) was estimated based on daily transport truck 451

trips and the AP-42 refueling losses emission factor (see Supplementary Materials) and subtract-452

ed from total evaporative fuel emission rate. 453

The mean and standard deviation for each hourly vehicle evaporative emission factor were 454

again estimated using the Monte-Carlo technique. 3000 simulations were undertaken by random-455

ly varying parking lots and the gas transport company location with a standard deviation of 3 456

grid cells around the actual location of these facilities to account for uncertainties in φ. In addi-457

tion, an uncertainty of 50% in the AP-42 truck refueling emission factor was assumed, and the 458

number of truck trips per hour was assumed to have an uncertainty of 25%. Half of the estimated 459

emission factors had a standard deviation of less than 17% and 90% of the data had a standard 460

deviation of less than 50%. The smaller uncertainty indicates that spatial variation of φ was rela-461

tively small near the source region. 10 samples (approximately 14%) were removed as outliers 462

from the analysis. Table 2 shows the statistical analysis of the hourly data. The mean vehicle 463

evaporative emission rate was found to be 2.9 g h-1

vehicle-1

; the median emission rate was 2.3 g 464

h-1

vehicle -1

, with a 95% confidence interval of [1.3, 3.3] g h-1

vehicle-1

.

465

The evaporative fuel emission factor from the flux-footprint analysis was again compared 466

with MOVES estimated values. The YCT site features a constant turnover of taxi cabs during 467

daytime coming to and from headquarters, stopping or parking short-term near the tower, which 468

is expected to contribute additional hot-soak emissions. The MOVES based average daytime 469

emission factor for the parked vehicles (assuming half of the vehicles with peak hot soak emis-470

sions and the remaining half with average hot soak emissions) were 0.41 g h-1

vehicle-1

and 0.55 471

g h-1

vehicle-1

for year 2008 and 2000 vehicle fleets, respectively, which is approximately 18% 472

and 24% of the median emission rate from the flux-footprint analysis. Thus, uncertainty in vehi-473

cle ages cannot explain the discrepancy and alternative explanations were explored. 474

Uncertainties in the estimation of number of parked vehicles in other areas in the footprint 475

domain and emissions from two fuel service stations (gas stations), located at 480 m NNE and 476

1380 m SE of the tower, as shown in Figure 1, were determined (see Supplementary Materials) 477

17

to have very small effects on the evaporative fuel emission factor. Another potential source of 478

evaporative emission is on-road vehicles. Based on vehicle density described in Section 3.2.2 479

and additional evaporative emission factors of running vehicles estimated using the MOVES 480

model (approximately 0.06 g km-1

at 48 kmh-1

), a decrease of the evaporative fuel emission fac-481

tor by approximately 10% was obtained, which still cannot explain the significant discrepancy 482

between the obtained vehicle fuel emission factors based on the flux source apportionment data 483

and the MOVES model. Possible explanations for the difference between flux-based and 484

MOVES-based evaporative emission rates could be (i) evaporative fuel emissions from a large 485

auto-repair workshop 100 m east to the tower, (ii) a few percent of poorly maintained vehicles 486

that could have evaporative emission rates hundreds of times higher than well-maintained vehi-487

cles, and/or (iii) a significant underestimation of evaporative emissions by the MOVES model in 488

this environment, similar to previous results (Quigley, 2007). A more detailed inspection of the 489

vehicle fleet condition would be needed to confirm the existence of the first two possibilities. 490

491

4. Conclusions 492

In this study, a receptor-oriented statistical model and an analytical flux-footprint model were 493

utilized to analyze VOC flux data obtained from an urban area in Houston to determine the con-494

tributions of responsible sources of VOCs to observed flux and the VOC emission rates from 495

these sources. Emission rates from a foam plastics industry source, running vehicle exhaust and 496

evaporative emissions were calculated. Median VOC emissions from the industrial source were 497

15.7±3.1 kg h-1

, higher than officially permitted amounts, but potentially biased due to the large 498

distance from the source and higher temperatures in the summer months. Estimated vehicle ex-499

haust emissions, with a median emission rate of 14.5±2 g h-1

vehicle-1

, were similar to the esti-500

mates using the MOVES model and a vehicle fleet of year 2005 (14.1±0.2 g h-1

vehicle-1

), possi-501

bly representative of the vehicles used in the study domain. And finally, estimated evaporative 502

emissions from parked vehicles, with a median emission rate of 2.3±1 g h-1

vehicle-1

, were sig-503

nificantly higher than the MOVES model predictions, suggesting either (i) the existence of poor-504

ly maintained vehicles with much higher evaporative emissions, (ii) other sources apart from ve-505

hicles contributing to the evaporative fuel emissions flux., and/or (iii) a significant underestima-506

tion of evaporative fluxes by MOVES. Based on this study, while the reported evaporative emis-507

18

sion rates should be considered as an upper limit for parked vehicles in this area, more studies on 508

evaporative fuel emissions appear to be needed to validate the accuracy of the emissions model. 509

The flux methodology used in this study can be used as an alternative approach to measure 510

emission rates from sources for which direct emissions measurements are difficult or impossible. 511

For example, if deployed to a tall tower downwind of petrochemical industry regions in the Hou-512

ston Ship Channel, it could be used to estimate emissions from these industrial sources, including 513

fugitive and transient emissions, which are usually not reported accurately (Vizuete et al., 2008). 514

The method used here could also be used more widely to determine real-world emissions from 515

in-use vehicles and compare with estimations from emission factor models, with a goal of valida-516

tion. The advantage of this method is that it naturally estimates the emission rate under real 517

world driving and dilution conditions rather than under an artificial driving cycle and dilution 518

ratio in typical vehicle emission testing. It can also provide more details on the chemical compo-519

sition than remote sensing, which is limited in its ability in resolving chemical compositions 520

(Singer et al., 1998). However, as demonstrated in this study, more details regarding the vehicle 521

fleet composition, speeds and density are needed to improve the top-down versus bottom-up 522

emissions comparison, and to effectively validate and improve vehicle emission factor models. 523

524

Acknowledgements 525

Although the research described in the article has been funded in part by the U.S. Environmental 526

Protection Agency's Science to Achieve Results (STAR) program through grant (R834556), it 527

has not been subjected to any EPA review and therefore does not necessarily reflect the views of 528

the Agency, and no official endorsement should be inferred. 529

19

Table 1 List of measured VOCs and their abbreviations 530

IUPAC name (common name) Abbreviation Method Detection Limit (MDL), ppt

butane* C4 12.0

2-methyl 1,3-butandiene (isoprene) C5H8 10.0

n-pentane C5H12 10.0

2-methylbutane (isopentane) IC5H12 10.0

benzene BENZ 8.0

ethylbenzene EBENZ 6.0

n-hexane NC6H14 8.0

2-methylpentane M2PEN 8.0

3-methylpentane M3PEN 8.0

methylbenzene (toluene) TOLU 7.0

n-heptane NC7H12 7.0

2-methylhexane (isoheptane) M2HEX 7.0

2,2-dimethylpentane (neoheptane) NEOH 7.0

1,3- and 1,2-dimethylbenzene (m/p-xylene) MPXYL 6.0

1,2-dimethylbenzene (o-xylene) OXYL 6.0

2-methylprop-2-enal (methacrolein) MACR 16.0

butenone (methyl vinyl ketone) MVK 16.0

butanone (methyl ethyl ketone) MEK 16.0

* Note: includes n-butane and 2-methylpropane (isobutane). 531

20

Table 2 Descriptive statistics for the hourly emission rates 532

Foam Plastics

Industry source

Vehicle

exhaust emissions

Vehicle

evaporate emissions

(kg h-1

) (g h-1

vehicle-1

) (g h-1

vehicle-1

)

Number of data points 107 204 60

Minimum 0.2 0.1 0.1

1st quartile 7.6 5.7 0.5

Median 15.7 14.5 2.3

3rd quartile 27.5 23.9 5.3

Maximum 53.3 55.1 11.4

Lower 95% confidence limit for median 12.6 12.5 1.4

Upper 95% confidence limit for median 18.8 16.5 3.3

Mean 18.5 17.5 2.9

Standard deviation 13.6 14.2 0.8

Skewness 0.7 0.9 0.7

533

534

21

535

536

537

Figure 1: A schematic showing the positions of the Yellow Cab Tower (YCT), nearby major 538

roadways (Collingsworth St., Quitman St., Hardy St., and Elysian St.), surface parking lots near 539

YCT, a foam plastics industry site, a gasoline transport refilling facility and two refueling sta-540

tions. Numbers on the x and y axes represent distance in m from the origin of the flux footprint 541

model domain. See text and Figure 6 for more details. 542

543

22

544

Figure 2: Predicted source profiles (mg mg-1

) by ME-2 based on the flux data. Error bars are 545

standard deviations estimated using bootstrap analyses.546

23

547 548

549

Figure 3: Observed and reconstructed fluxes of VOC species measured at the Yellow Cab Tow-550

er. Thin solid lines represent 1:1, 1:2 and 2:1 ratios. Units are mg m-2

h-1

. Note that the data 551

points with missing observations replaced by median values are not shown in the plot. 552

553

24

554

Figure 4: ME-2 predicted average hourly source contributions (mg m-2

h-1

) to the measured VOC 555

fluxes at the Yellow Cab Tower. The box-and-whisker plot shows the median, min, max and in-556

terquartile range of the data for each hour. 95% confidence intervals of the median are shown in 557

red. 558 559

25

560

561

Figure 5: Wind direction dependence of the ME-2 apportioned fluxes of measured VOCs: (a) 562

consumer and solvent use emissions, (b) foam plastics industry emissions, (c) vehicle exhaust, (d) 563

evaporative fuel emissions, and (e) biogenic emissions. Units are mg m-2

h-1

. Negative fluxes are 564

shown in red. 565

26

566

567 568

Figure 6: Averaged footprint function in the domain for (a) 0700-1200 and (b) 1300-1900 CST. 569

Numbers on the x and y axes are distance in m. Maximum values are approximately 3.35×10-3

570

for grid cells close to the tower, which is located at center (3000 m, 3000 m). See Figure 1 for 571

details of the major roadways and other emission sources. 572

573

27

574 Figure 7: Comparison of vehicle exhaust emission rates estimated using the flux-footprint analy-575

sis and the MOVES model. Uncertainties of the MOVES emission are one standard deviation 576

about the mean, estimated using a Monte-Carlo technique that considers the uncertainties in the 577

vehicle volume and speed, as used in the flux-footprint calculations. 578

28

References 579

Amato, F., Pandolfi, M., Escrig, A., Querol, X., Alastuey, A., Pey, J., Perez, N., Hopke, P.K., 580

2009. Quantifying road dust resuspension in urban environment by Multilinear Engine: A 581

comparison with PMF2. Atmospheric Environment 43, 2770-2780. 582

Atkinson, R., 2000. Atmospheric chemistry of VOCs and NOx. Atmospheric Environment 34, 583

2063-2101. 584

Brown, S.G., Reid, S.B., Roberts, P.T., Buhr, M.P., Funk, T.H., Kim, E., Hopke, P.K., 2004. 585

Reconciliation of the VOC and NOx Emission Inventory with Ambient Data in the Houston, 586

Texas Region, 13th International Emission Inventory Conference "Working for Clean Air in 587

Clearwater", Clearwater, FL. 588

Buzcu, B., Fraser, M.P., 2006. Source identification and apportionment of volatile organic 589

compounds in Houston, TX. Atmospheric Environment 40, 2385-2400. 590

Foken, T., Wichura, B., 1996. Tools for quality assessment of surface-based flux measurements. 591

Agricultural and Forest Meteorology 78, 83-105. 592

Fujita, E.M., Lu, Z., Sagebiel, J., Robinson, N.F., Watson, J.G., 1995. VOC source 593

apportionment for the Coast Oxidant Assessment for Southeast Texas. Texas Natural Resource 594

Conservation Commission,Desert Research Institute, Austin,TX. 595

Gilman, J.B., Kuster, W.C., Goldan, P.D., Herndon, S.C., Zahniser, M.S., Tucker, S.C., Brewer, 596

W.A., Lerner, B.M., Williams, E.J., Harley, R.A., Fehsenfeld, F.C., Warneke, C., de Gouw, J.A., 597

2009. Measurements of volatile organic compounds during the 2006 TexAQS/GoMACCS 598

campaign: Industrial influences, regional characteristics, and diurnal dependencies of the OH 599

reactivity. Journal of Geophysical Research-Atmospheres 114. 600

Gustafson, W.I., Yu, S., 2012. Generalized approach for using unbiased symmetric metrics with 601

negative values: normalized mean bias factor and normalized mean absolute error factor. 602

Atmospheric Science Letters 13, 262-267. 603

Henry, R.C., Spiegelman, C.H., Collins, J.F., Park, E., 1997. Reported emissions of organic 604

gases are not consistent with observations. Proceedings of the National Academy of Sciences of 605

the United States of America 94, 6596-6599. 606

Hsu, Y., Divita, F., 2008. SPECIATE 4.2 Speciation database development documentation. Draft 607

Report. Prepared for Office of Research and Development, U.S. Environmental Protection 608

Agency. E.H. Pechan & Associates, Inc. 609

Karl, T., Apel, E., Hodzic, A., Riemer, D.D., Blake, D.R., Wiedinmyer, C., 2009. Emissions of 610

volatile organic compounds inferred from airborne flux measurements over a megacity. 611

Atmospheric Chemistry and Physics 9, 271-285. 612

Karl, T., Guenther, A., Jordan, A., Fall, R., Lindinger, W., 2001. Eddy covariance measurement 613

of biogenic oxygenated VOC emissions from hay harvesting. Atmospheric Environment 35, 491-614

495. 615

Kim, E., Brown, S.G., Hafner, H.R., Hopke, P.K., 2005. Characterization of non-methane 616

volatile organic compounds sources in Houston during 2001 using positive matrix factorization. 617

Atmospheric Environment 39, 5934-5946. 618

Kim, E., Hopke, P.K., Paatero, P., Edgerton, E.S., 2003. Incorporation of parametric factors into 619

multilinear receptor model studies of Atlanta aerosol. Atmospheric Environment 37, 5009-5021. 620

Kim, S.W., McKeen, S.A., Frost, G.J., Lee, S.H., Trainer, M., Richter, A., Angevine, W.M., 621

Atlas, E., Bianco, L., Boersma, K.F., Brioude, J., Burrows, J.P., de Gouw, J., Fried, A., Gleason, 622

J., Hilboll, A., Mellqvist, J., Peischl, J., Richter, D., Rivera, C., Ryerson, T., Hekkert, S.T.L., 623

Walega, J., Warneke, C., Weibring, P., Williams, E., 2011. Evaluations of NOx and highly 624

29

reactive VOC emission inventories in Texas and their implications for ozone plume simulations 625

during the Texas Air Quality Study 2006. Atmospheric Chemistry and Physics 11, 11361-11386. 626

Kljun, N., Calanca, P., Rotach, M., Schmid, H., 2004. A Simple Parameterisation for Flux 627

Footprint Predictions. Boundary-Layer Meteorology 112, 503-523. 628

Kljun, N., Kormann, R., Rotach, M.W., Meixer, F.X., 2003. Comparison of the Langrangian 629

footprint model LPDM-B with an analytical footprint model. Boundary-Layer Meteorology 106, 630

349-355. 631

Kljun, N., Rotach, M.W., Schmid, H.P., 2002. A Three-Dimensional Backward Lagrangian 632

Footprint Model For A Wide Range Of Boundary-Layer Stratifications. Boundary-Layer 633

Meteorology 103, 205-226. 634

Kormann, R., Meixner, F., 2001. An Analytical Footprint Model For Non-Neutral Stratification. 635

Boundary-Layer Meteorology 99, 207-224. 636

Kota, S.H., Ying, Q., Zhang, Y., 2013. Simulating near-road reactive dispersion of gaseous air 637

pollutants using a three-dimensional Eulerian model. Science of the Total Environment 454-455, 638

348-357. 639

Kroll, J.H., Seinfeld, J.H., 2008. Chemistry of secondary organic aerosol: Formation and 640

evolution of low-volatility organics in the atmosphere. Atmospheric Environment 42, 3593-3624. 641

Langford, B., Davison, B., Nemitz, E., Hewitt, C.N., 2009. Mixing ratios and eddy covariance 642

flux measurements of volatile organic compounds from an urban canopy (Manchester, UK). 643

Atmospheric Chemistry and Physics 9, 1971-1987. 644

Langford, B., Misztal, P.K., Nemitz, E., Davison, B., Helfter, C., Pugh, T.A.M., MacKenzie, 645

A.R., Lim, S.F., Hewitt, C.N., 2010a. Fluxes and concentrations of volatile organic compounds 646

from a South-East Asian tropical rainforest. Atmospheric Chemistry and Physics 10, 8391-8412. 647

Langford, B., Nemitz, E., House, E., Phillips, G.J., Famulari, D., Davison, B., Hopkins, J.R., 648

Lewis, A.C., Hewitt, C.N., 2010b. Fluxes and concentrations of volatile organic compounds 649

above central London, UK. Atmospheric Chemistry and Physics 10, 627-645. 650

Lee, A., Schade, G.W., Holzinger, R., Goldstein, A.H., 2005. A comparison of new 651

measurements of total monoterpene flux with improved measurements of speciated monoterpene 652

flux. Atmospheric Chemistry and Physics 5, 505-513. 653

Leuchner, M., Rappenglück, B., 2010. VOC source–receptor relationships in Houston during 654

TexAQS-II. Atmospheric Environment 44, 4056-4067. 655

McKeen, S., Grell, G., Peckham, S., Wilczak, J., Djalalova, I., Hsie, E.Y., Frost, G., Peischl, J., 656

Schwarz, J., Spackman, R., Holloway, J., de Gouw, J., Warneke, C., Gong, W., Bouchet, V., 657

Gaudreault, S., Racine, J., McHenry, J., McQueen, J., Lee, P., Tang, Y., Carmichael, G.R., 658

Mathur, R., 2009. An evaluation of real-time air quality forecasts and their urban emissions over 659

eastern Texas during the summer of 2006 Second Texas Air Quality Study field study. Journal of 660

Geophysical Research-Atmospheres 114. 661

Mills, N.J., 2007. Chapter 3 - Foamed thermoplastics: Microstructure and processing, Polymer 662

Foams Handbook. Butterworth-Heinemann, Oxford, pp. 39-67. 663

Moore, D.S., McCabe, G.P., 1999. Introduction to the Pratice of Statistics, 3rd ed. W. H. 664

Freeman, New York. 665

Nam, J., Kimura, Y., Vizuete, W., Murphy, C., Allen, D.T., 2006. Modeling the impacts of 666

emission events on ozone formation in Houston, Texas. Atmospheric Environment 40, 5329-667

5341. 668

30

Norris, G.A., R. Vedantham, K. Wade, P. Zhan, S. Brown, P. Pentti, S. I. Eberly, and C. Foley, 669

2009. Guidance Document for PMF Applications with the Multilinear Engine. U.S. 670

Environmental Protection Agency, Washington, DC. 671

Paatero, P., 1999. The Multilinear Engine: A Table-Driven, Least Squares Program for Solving 672

Multilinear Problems, including the n-Way Parallel Factor Analysis Model. Journal of 673

Computational and Graphical Statistics 8, 854-888. 674

Paatero, P., 2004. End user's guide to multilinear engine applications. 675

Park, C., Schade, G.W., Boedeker, I., 2010. Flux measurements of volatile organic compounds 676

by the relaxed eddy accumulation method combined with a GC-FID system in urban Houston, 677

Texas. Atmospheric Environment 44, 2605-2614. 678

Park, C., Schade, G.W., Boedeker, I., 2011. Characteristics of the flux of isoprene and its 679

oxidation products in an urban area. J. Geophys. Res. 116, D21303. 680

Polissar, A.V., Hopke, P.K., Paatero, P., Malm, W.C., Sisler, J.F., 1998. Atmospheric aerosol 681

over Alaska: 2. Elemental composition and sources. Journal of Geophysical Research: 682

Atmospheres 103, 19045-19057. 683

Polissar, A.V., Hopke, P.K., Poirot, R.L., 2001. Atmospheric aerosol over Vermont: Chemical 684

composition and sources. Environmental Science & Technology 35, 4604-4621. 685

Quigley, C.J., 2007. Refueling and Evaporative Emissions of Volatile Organic Compounds from 686

Gasoline Powered Motor Vehicles, Civil Engineering. University of Texas, Austin, p. 198. 687

Ramadan, Z., Eickhout, B., Song, X.-H., Buydens, L.M.C., Hopke, P.K., 2003. Comparison of 688

Positive Matrix Factorization and Multilinear Engine for the source apportionment of particulate 689

pollutants. Chemometrics and Intelligent Laboratory Systems 66, 15-28. 690

Reid, S.B., Chinkin, L.R., Penfold, B.M., Gilliland, E.K., 2000. Emissions Inventory Validation 691

and Improvement: A Central California Case Study. Sonama Technology, Inc., Petaluma, CA. 692

Rubin, J.I., Kean, A.J., Harley, R.A., Millet, D.B., Goldstein, A.H., 2006. Temperature 693

dependence of volatile organic compound evaporative emissions from motor vehicles. Journal of 694

Geophysical Research: Atmospheres 111, D03305. 695

Singer, B.C., Harley, R.A., Littlejohn, D., Ho, J., Vo, T., 1998. Scaling of Infrared Remote 696

Sensor Hydrocarbon Measurements for Motor Vehicle Emission Inventory Calculations. 697

Environmental Science & Technology 32, 3241-3248. 698

Spirig, C., Neftel, A., Ammann, C., Dommen, J., Grabmer, W., Thielmann, A., Schaub, A., 699

Beauchamp, J., Wisthaler, A., Hansel, A., 2005. Eddy covariance flux measurements of biogenic 700

VOCs during ECHO 2003 using proton transfer reaction mass spectrometry. Atmospheric 701

Chemistry and Physics 5, 465-481. 702

USEPA, 2008. EPA Positive Matrix Factorization (PMF) 3.0 Fundamentals and User Guide, 703

Research Triangle Park, NC. 704

Velasco, E., Pressley, S., Grivicke, R., Allwine, E., Coons, T., Foster, W., Jobson, B.T., 705

Westberg, H., Ramos, R., Hernandez, F., Molina, L.T., Lamb, B., 2009. Eddy covariance flux 706

measurements of pollutant gases in urban Mexico City. Atmospheric Chemistry and Physics 9, 707

7325-7342. 708

Vizuete, W., Kim, B.U., Jeffries, H., Kimura, Y., Allen, D.T., Kioumourtzoglou, M.A., Biton, L., 709

Hendersona, B., 2008. Modeling ozone formation from industrial emission events in Houston, 710

Texas. Atmospheric Environment 42, 7641-7650. 711

Wu, C.-f., Larson, T.V., Wu, S.-y., Williamson, J., Westberg, H.H., Liu, L.J.S., 2007. Source 712

apportionment of PM2.5 and selected hazardous air pollutants in Seattle. Science of the Total 713

Environment 386, 42-52. 714

31

Ying, Q., Krishnan, A., 2010. Source contributions of volatile organic compounds to ozone 715

formation in southeast Texas. Journal of Geophysical Research-Atmospheres 115, D17306. 716

717 718